Surface Defect Detection Of Industrial Steel Strip Based On Machine Vision

Steel strip is an industrial precision product made of carbon steel as raw material,commonly used for pulling components and bundling goods.Its quality is closely related to industrial safety,and defect detection is an important barrier to prevent its defective products from entering the market.At present,defect detection based on machine vision mainly relies on artificially designed image features,which cannot achieve universality of multiple types of defects.Moreover,the accuracy and robustness of the model are poor,and it has a high error rate,making it difficult to meet the production needs of modern factory automation production lines.This article fully investigates the mechanism of industrial steel strip image defects and the theory of machine vision related algorithms,aiming to improve the performance of the detection model and make it more suitable for industrial production.Therefore,this article conducts research based on actual industrial steel strip defect images,and the main research content is as follows:(1)This article analyzes the characteristics of industrial steel strip data images,conducts research on different defect types of steel strips,and proposes a surface defect detection model for industrial steel strips based on improved YOLO V5.In view of the high proportion of small defects in industrial steel strip images and the difficulty in extracting feature information,this paper introduces SE(Squeeze and Extraction)attention mechanism in YOLO V5 S backbone network to strengthen the model’s ability to extract important information about defects and solve the problem of large loss of feature information;On this basis,the YOLO feature pyramid structure is analyzed,and four scale feature extraction is realized by increasing the detection area of the network to strengthen the integration of deep and shallow Semantic information of the model;And analyze the steel strip image through k-means++algorithm,redesign the size of the model anchor frame,and solve the problem of difficult detection of some small targets;Finally,experimental comparisons were conducted on the industrial steel strip dataset,and the results showed that the improved model effectively improved the detection accuracy of small targets while ensuring detection speed.(2)To further improve the accuracy and efficiency of defect detection models and adapt them to the needs of precision factory preparation,this article integrates the Swin Transformer module for further optimization.Firstly,a sliding attention mechanism is introduced,which allows information to be transmitted between adjacent windows through sliding window interaction,achieving a global modeling effect;Then a new upsampling operator is used to make the model have a larger Receptive field,which can make better use of the surrounding information,and the amount of calculation parameters is less;At the same time,in order to ensure the detection effect of the detection network,Adaptive Spatial Feature Fusion(ASFF)is also introduced,which weights the three horizontal feature maps output by the feature pyramid structure for feature fusion.By fully utilizing features at different scales,adaptive learning parameters are set to suppress the inconsistency caused by gradient backpropagation during the training process of the model;The final results on the dataset show that the model improved the average accuracy of the mean by 7 percentage points,achieving a high defect detection effect at the expense of less detection speed.In summary,this article takes industrial steel strips as the research objective,conducts research work based on machine vision,and designs corresponding detection models for different application scenarios,which has certain theoretical value and practical significance.